Add part of AVX SSE4.1 packed move with sign/zero extend instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@107720 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 112 insertions, 0 deletions

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 66486e9df8..7c678a67d4 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -4304,6 +4304,21 @@ multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
        OpSize;
 }
 
+let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
+defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
+                                     VEX;
+defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
+                                     VEX;
+defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
+                                     VEX;
+defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
+                                     VEX;
+defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
+                                     VEX;
+defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
+                                     VEX;
+}
+
 defm PMOVSXBW   : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
 defm PMOVSXWD   : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
 defm PMOVSXDQ   : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
diff --git a/test/MC/AsmParser/X86/x86_32-encoding.s b/test/MC/AsmParser/X86/x86_32-encoding.s
index bc767d15d4..34792a901f 100644
--- a/test/MC/AsmParser/X86/x86_32-encoding.s
+++ b/test/MC/AsmParser/X86/x86_32-encoding.s
@@ -11950,3 +11950,51 @@
 // CHECK: encoding: [0xc4,0xe3,0x71,0x4c,0x18,0x20]
           vpblendvb  %xmm2, (%eax), %xmm1, %xmm3
 
+// CHECK: vpmovsxbw  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x20,0xea]
+          vpmovsxbw  %xmm2, %xmm5
+
+// CHECK: vpmovsxbw  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x20,0x10]
+          vpmovsxbw  (%eax), %xmm2
+
+// CHECK: vpmovsxwd  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x23,0xea]
+          vpmovsxwd  %xmm2, %xmm5
+
+// CHECK: vpmovsxwd  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x23,0x10]
+          vpmovsxwd  (%eax), %xmm2
+
+// CHECK: vpmovsxdq  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x25,0xea]
+          vpmovsxdq  %xmm2, %xmm5
+
+// CHECK: vpmovsxdq  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x25,0x10]
+          vpmovsxdq  (%eax), %xmm2
+
+// CHECK: vpmovzxbw  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x30,0xea]
+          vpmovzxbw  %xmm2, %xmm5
+
+// CHECK: vpmovzxbw  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x30,0x10]
+          vpmovzxbw  (%eax), %xmm2
+
+// CHECK: vpmovzxwd  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x33,0xea]
+          vpmovzxwd  %xmm2, %xmm5
+
+// CHECK: vpmovzxwd  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x33,0x10]
+          vpmovzxwd  (%eax), %xmm2
+
+// CHECK: vpmovzxdq  %xmm2, %xmm5
+// CHECK: encoding: [0xc4,0xe2,0x79,0x35,0xea]
+          vpmovzxdq  %xmm2, %xmm5
+
+// CHECK: vpmovzxdq  (%eax), %xmm2
+// CHECK: encoding: [0xc4,0xe2,0x79,0x35,0x10]
+          vpmovzxdq  (%eax), %xmm2
+
diff --git a/test/MC/AsmParser/X86/x86_64-encoding.s b/test/MC/AsmParser/X86/x86_64-encoding.s
index 89701503cd..ef274cf0a4 100644
--- a/test/MC/AsmParser/X86/x86_64-encoding.s
+++ b/test/MC/AsmParser/X86/x86_64-encoding.s
@@ -1998,3 +1998,52 @@ pshufb	CPI1_0(%rip), %xmm1
 // CHECK: encoding: [0xc4,0x63,0x21,0x4c,0x28,0xc0]
           vpblendvb  %xmm12, (%rax), %xmm11, %xmm13
 
+// CHECK: vpmovsxbw  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x20,0xd4]
+          vpmovsxbw  %xmm12, %xmm10
+
+// CHECK: vpmovsxbw  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x20,0x20]
+          vpmovsxbw  (%rax), %xmm12
+
+// CHECK: vpmovsxwd  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x23,0xd4]
+          vpmovsxwd  %xmm12, %xmm10
+
+// CHECK: vpmovsxwd  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x23,0x20]
+          vpmovsxwd  (%rax), %xmm12
+
+// CHECK: vpmovsxdq  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x25,0xd4]
+          vpmovsxdq  %xmm12, %xmm10
+
+// CHECK: vpmovsxdq  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x25,0x20]
+          vpmovsxdq  (%rax), %xmm12
+
+// CHECK: vpmovzxbw  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x30,0xd4]
+          vpmovzxbw  %xmm12, %xmm10
+
+// CHECK: vpmovzxbw  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x30,0x20]
+          vpmovzxbw  (%rax), %xmm12
+
+// CHECK: vpmovzxwd  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x33,0xd4]
+          vpmovzxwd  %xmm12, %xmm10
+
+// CHECK: vpmovzxwd  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x33,0x20]
+          vpmovzxwd  (%rax), %xmm12
+
+// CHECK: vpmovzxdq  %xmm12, %xmm10
+// CHECK: encoding: [0xc4,0x42,0x79,0x35,0xd4]
+          vpmovzxdq  %xmm12, %xmm10
+
+// CHECK: vpmovzxdq  (%rax), %xmm12
+// CHECK: encoding: [0xc4,0x62,0x79,0x35,0x20]
+          vpmovzxdq  (%rax), %xmm12
+
+